a web application firewall for nginx - appsec · thibault koechlin (nbs-system) why naxsi web...

A web application firewall for nginx

Thibault Koechlin (nbs-system)

Why Naxsi

Web application Classical IT

Why naxsiBest mitigation : patch

Not always possible :• Lack of skill• Application is too « critical »

Why naxsi

Fallback : WAFs ? (back in 2011 \o/)

Commercial Wafs :• Very uneaven• Not really affordable for small companies with

big infrastructures

Opensource WAFs :• No waf at that time for nginx• Not found of complex signatures:)

Why naxsiAs a pentester :• Web application still the most vulnerable and

exposed part of the perimeterAs a hoster :• Website owners, even when web is at the core

of business, lacks awareness … and get owned

As a security « consultant » :• CISO/Admin still frightened of side effects• Open-source WAFs only seen in very « tech

savy » companies

Why u no protect ?!

Introducing naxsi

Naxsi – yet another WAF ?• Rather than detecting « complex » signatures, will focus on « tokens » :

– <>()[];='#...

• Tokens presence leads to score increase → action

• Small code base :

– core (~4k) C

– learning tools (~1.3k) python

Integrates libinjection as well !

Workflow• Generated logs are

pushed into ES• Data is relevant for

monitoring and trafic inspection

• Nxapi helps the admin generate whitelists

Strengths & Weaknesses

Strength WeaknessesGood resilience against

unknown/obfuscated attacksInitial learning is needed

Good performances (low mem footprint, minimal runtime processing)

Fast evolving apps requires coordination with releases

No need for updates of « attack » signatures

No « intelligence », unsuitable for some specific cases

Learning process strongly assisted

Json + x-www-form-urlencoded + multipart/form-data

What does it actually looks like

• Detection rule :

• CheckRule :

• Whitelist :ARGSURLHEADERSBODY / FILE_EXT|NAME

MainRule "str:(" "mz:ARGS|URL|BODY|$HEADERS_VAR:cookie" "s:$SQL:4,$XSS:8" id:1010;

CheckRule "$SQL >= 8" BLOCK; #(DROP|LOG)

 BasicRule wl:1010 "mz:$HEADERS_VAR:cookie|$URL:/x" ;

Learning & tools

Learning process

• Relying on (ES) injected logs• Suggests whitelists (nxapi) :

– Based on templates (application specific)– Statistics (number of occurrences, number of peers …)

• Associated events are then « tagged » into database

Learning process

• Naxsi relies on two main modes of operation « learning » / « blocking »– During learning phases, exceptions are

loggued but not blocked– Once learning is over, naxsi can be set to

blocking mode (bad trafic is dropped)

More into learning

Learning is the biggest downside, however :

• When it comes to « market » apps, whitelists are very predictable (templates!)

• For home-made apps, several options :– Relying on statistics– Relying on « trusted » trafic

More into learning

{ "_msg" : "Magento checkout page (BODY|NAME)", "?uri" : "/checkout/onepage/.*", "zone" : "BODY|NAME", "id" : "1310 OR 1311"}

BasicRule wl:1310,1311 "mz:$URL_X:^/checkout/onepage/savebilling/$|BODY|NAME";

Nxtool templates (dnyamic)

Naxsi templates (static)

More into reporting/visualisation

• <insert cool kibana dashboard cap here>

Tips & Tricks

Tweaks around learningCombined with nginx scripting :

• Learning only for some specifics URI(s)• Learning only from some IP(s)• Learning if the visitor fits some critetria

Naxsi on/offLearning on/offpost_action on/offextensive_log on/offLibinjection (sql|xss) on/off

if ($remote_addr = "1.2.3.4") { set $naxsi_flag_learning 0; set $naxsi_libinjection_sql 1 ; }

Tweaks around learning

Naxsi can be instructed to drop requests despite learning mode :• Libinjection

CheckRule "$LIBINJECTION_XSS >= 8" DROP;

CheckRule "$LIBINJECTION_SQL >= 8" DROP;

• Doxi-rules http://spike.nginx-goodies.com/rules/

CheckRule "$UWA >= 8" DROP;

Feedback from real life :

• Rules syntax stay very simple :– Lowers the risk of breach while playing

around rules– Lowers the needed skill to manage the rules

• Naxsi itself is very simple :– Low ressources– Specific cases might become problematic :

content legitimately passed in base64

Achievements• As been tested in several occasions :

– Real life (a lot)– Audited by 3rd parties– Challenges

• Used to protect some website under « persistent » attacks

• Used to protect at least one very large website (1Tb+)

What's next,Q&A

Thanks for your attention !